CN111414860A - Real-time portrait tracking and segmenting method - Google Patents
Real-time portrait tracking and segmenting method Download PDFInfo
- Publication number
- CN111414860A CN111414860A CN202010200344.2A CN202010200344A CN111414860A CN 111414860 A CN111414860 A CN 111414860A CN 202010200344 A CN202010200344 A CN 202010200344A CN 111414860 A CN111414860 A CN 111414860A
- Authority
- CN
- China
- Prior art keywords
- portrait
- frame
- tracking
- mask
- segmentation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000011218 segmentation Effects 0.000 claims abstract description 39
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000003709 image segmentation Methods 0.000 claims abstract description 7
- 238000012545 processing Methods 0.000 claims description 11
- 230000007812 deficiency Effects 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012805 post-processing Methods 0.000 claims description 4
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 8
- 238000002372 labelling Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/162—Detection; Localisation; Normalisation using pixel segmentation or colour matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a real-time portrait tracking and segmenting method. The method specifically comprises the following steps: (1) the training stage is used for training parameters of the human image segmentation model and performing off-line training on the segmentation network, and comprises two parts, namely data preprocessing and model training; (2) in the prediction stage, picture sequence frames of a video are input into a portrait tracking segmentation algorithm, a portrait tracking frame is obtained through a KCF tracking algorithm, a portrait area is cut out according to the portrait tracking frame, the portrait area is preprocessed and input into a segmentation model, an output result is post-processed, a portrait segmentation mask corresponding to an input frame is obtained, the process is circulated according to the sequence of the video frames until a final video portrait mask sequence is obtained, and the process is finished. The invention has the beneficial effects that: the running speed of the algorithm is improved; the picture data is easier to obtain and label; the running speed of the model is increased, and the real-time requirement of the mobile terminal is met.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a real-time portrait tracking and segmenting method.
Background
Tracking algorithms and segmentation algorithms belong to two different technical fields. Generally, in a scene, a tracking algorithm mainly continuously tracks a given target so as to obtain the position information of the target in the scene. The segmentation algorithm is mainly used for performing semantic segmentation on a given target so as to obtain a series of target masks. Therefore, if the tracking algorithm and the segmentation algorithm are combined together, the application of the tracking algorithm and the segmentation algorithm is quite wide, for example, the tracking segmentation algorithm of the portrait is used in the short video industry, and a basis can be provided for playing methods such as rendering special effects of videos and the like.
Most of the current tracking segmentation algorithms are based on deep learning methods. In terms of data, due to the need to train the tracking and the segmentation networks at the same time, video must be used as training data, and the labeling of video data takes a lot of manual time. The labeling of video data becomes a difficult point. In the aspect of models, due to the complexity of tracking and segmenting algorithms, the complexity of model structures is increased, the operation time of the algorithms is long, and the real-time requirements cannot be met.
Disclosure of Invention
The invention provides a real-time portrait tracking and segmenting method for improving the running speed of the algorithm in order to overcome the defects in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a real-time portrait tracking and segmenting method specifically comprises the following steps:
(1) the training stage is used for training parameters of the human image segmentation model and performing off-line training on the segmentation network, and comprises two parts, namely data preprocessing and model training;
(2) in the prediction stage, picture sequence frames of a video are input into a portrait tracking segmentation algorithm, a portrait tracking frame is obtained through a KCF tracking algorithm, a portrait area is cut out according to the portrait tracking frame, the portrait area is preprocessed and input into a segmentation model, an output result is post-processed, a portrait segmentation mask corresponding to an input frame is obtained, the process is circulated according to the sequence of the video frames until a final video portrait mask sequence is obtained, and the process is finished.
The invention adopts the traditional KCF tracking algorithm, and improves the running speed of the algorithm; the picture is taken as training data of the segmentation network, and compared with video data, the picture data is easier to acquire and label; designing a lightweight segmentation network, improving the running speed of the model and meeting the real-time requirement of a mobile terminal; therefore, the invention improves the running speed of the whole algorithm, and the tracking segmentation algorithm can meet the real-time performance requirement at the mobile terminal.
Preferably, in the step (1), the specific operation method is as follows:
(11) collecting different portrait data, and accurately marking a portrait area, wherein the background area is 0, and the portrait area is 1, so as to obtain a corresponding binary portrait mask;
(12) performing data enhancement processing on the training portrait data, then scaling the long edge of the image to 224, scaling the short edge in an equal ratio, and compensating for the deficiency to be 0 to be aligned to obtain an RGB input image I with the size of 224X 3xCarrying out the same scaling operation on the corresponding binary portrait mask to obtain a training portrait mask Iy;
(13) The method comprises the steps that Mobilenetv2 is used as a coding module of a segmentation network, and the whole coding module carries out 32-time down-sampling on an input image to obtain a feature map F; in the decoding module, a decoding mode similar to U-net is adopted to recover details of the feature diagram F, and after the size of the feature diagram is recovered to 56 × 56, 4 times of upsampling layers are directly utilized to obtain an output portrait mask Y with the size of 224 × 1;
(14) for output portrait mask Y and training portrait mask IyPerforming cross entropy loss function operation and Dice loss function operation to obtain a loss function L oss;
(15) the entire model is iterated using the penalty function L oss, updating the model parameters.
Preferably, in step (12), the data enhancement process includes mirroring, rotation, luminance-contrast transformation, affine transformation.
Preferably, in the step (2), the specific operation method is as follows:
(21) performing frame dismantling processing on the video to obtain a picture sequence frame;
(22) inputting the first frame of picture into a KCF tracking algorithm, and manually marking a portrait frame needing to be tracked to initiateKCF tracking algorithm is changed and taken as portrait tracking frame B of the first frame1;
(23) Suppose the portrait tracking frame of the current frame is BtWhere t represents the position of the picture frame in the video, the KCF tracking algorithm tracks the frame B according to the portrait of the current frametTo predict the portrait tracking frame B of the next framet+1;
(24) Tracking the frame B according to the portrait obtained in the step (23)tSelf-adaptive clipping is carried out on the portrait area to obtain the portrait area Pt;
(25) For portrait region PtPreprocessing the portrait region PtScaling the long side to 224, scaling the short side equally, and supplementing 0 to the deficiency to obtain the RGB model input I with the size of 224X 3t;
(26) RGB model input ItObtaining a portrait mask output Y after the portrait segmentation model obtained in the step (1)t;
(27) Outputting Y to portrait masktPost-processing is carried out to optimize the result and obtain a binary portrait mask N corresponding to the original image framet;
(28) And (5) circulating the steps (23) to (27) until the image segmentation of the last frame is completed, and obtaining all portrait mask frames.
Preferably, in step (24), the adaptive clipping specifically includes: firstly, judging the width-to-height ratio or the height-to-width ratio of a portrait tracking frame, if the width-to-height ratio or the height-to-width ratio is less than 0.5, expanding the range of a short side to enable the ratio of the short side to reach 0.5; then, the width and the height of the portrait tracking frame are expanded in the same proportion to obtain a portrait area PtAnd the whole cutting area is ensured to contain a complete portrait.
Preferably, in step (27), the image is masked and Y is outputtThe post-treatment specifically comprises the following steps: firstly, carrying out binarization processing on the image to obtain a binarized portrait mask, wherein the threshold value is 0.5; secondly, performing communication domain analysis on the portrait mask, removing the mistakenly-segmented regions, and reserving the maximum region of the portrait; the result is then scaled to the portrait area P in step (24)tThe size of (a); finally according to the portraitRegion PtThe cutting information of the image mask is compensated with 0 to obtain a binary image mask N corresponding to the original image framet。
The invention has the beneficial effects that: the running speed of the algorithm is improved; the picture data is easier to obtain and label; the running speed of the model is increased, and the real-time requirement of a mobile terminal is met; the running speed of the whole algorithm is improved, so that the tracking segmentation algorithm can meet the real-time performance requirement at the mobile terminal.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a flow chart of the segmentation model of the present invention.
Detailed Description
The invention is further described with reference to the following figures and detailed description.
In the embodiment shown in fig. 1, a real-time human image tracking and segmenting method specifically includes the following steps:
(1) the method comprises the following steps that parameters of an image segmentation model are trained in a training stage, and since the KCF tracking algorithm does not need offline training, only offline training is carried out on a segmentation network in the step, and the step comprises two parts, namely data preprocessing and model training; the specific operation method comprises the following steps:
(11) collecting different portrait data, and accurately marking a portrait area, wherein the background area is 0, and the portrait area is 1, so as to obtain a corresponding binary portrait mask;
(12) in order to improve the generalization of the network, the training portrait data is subjected to data enhancement processing, the data enhancement processing comprises mirroring, rotation, brightness contrast transformation, affine transformation and the like, then the long edge of the image is scaled to 224, the short edge is scaled in an equal ratio, the deficiency is compensated with 0 to be aligned, and the RGB input image I with the size of 224 x 3 is obtainedxCarrying out the same scaling operation on the corresponding binary portrait mask to obtain a training portrait mask Iy;
(13) The method comprises the steps that Mobilenetv2 is used as a coding module of a segmentation network, the coding module mainly comprises a series of Mobilenetv2 units, and the whole coding module carries out 32-time down-sampling on an input image to obtain a feature map F; in the decoding module, a decoding mode similar to U-net is adopted to recover details of the feature diagram F, and in order to reduce the calculation amount, after the size of the feature diagram is recovered to 56 × 56, 4 times of upsampling layers are directly utilized to obtain an output portrait mask Y with the size of 224 × 1;
(14) for output portrait mask Y and training portrait mask IyPerforming cross entropy loss function operation and Dice loss function operation to obtain a loss function L oss;
(15) the entire model is iterated using the penalty function L oss, updating the model parameters.
(2) Inputting picture sequence frames of a video into a portrait tracking segmentation algorithm in a prediction stage, obtaining a portrait tracking frame through a KCF tracking algorithm, cutting out a portrait area according to the portrait tracking frame, preprocessing the portrait area and inputting the preprocessed portrait area into a segmentation model, post-processing an output result to obtain a portrait segmentation mask corresponding to an input frame, and circulating the process according to the sequence of the video frames until a final video portrait mask sequence is obtained; the specific operation method comprises the following steps:
(21) performing frame dismantling processing on the video to obtain a picture sequence frame;
(22) inputting the first frame picture into a KCF tracking algorithm, manually marking a portrait frame needing to be tracked to initialize the KCF tracking algorithm, and taking the portrait frame as a portrait tracking frame B of the first frame1;
(23) Suppose the portrait tracking frame of the current frame is BtWhere t denotes the position of the picture frame in the video (t ═ 1, 2, 3 …), the KCF tracking algorithm tracks the frame B from the portrait of the current frametTo predict the portrait tracking frame B of the next framet+1;
(24) Tracking the frame B according to the portrait obtained in the step (23)tSelf-adaptive clipping is carried out on the portrait area to obtain the portrait area Pt(ii) a The self-adaptive cutting specifically comprises the following steps: firstly, judging the width-to-height ratio or the height-to-width ratio of a portrait tracking frame, if the width-to-height ratio or the height-to-width ratio is less than 0.5, expanding the range of a short side to enable the ratio of the short side to reach 0.5; then, the width and the height of the portrait tracking frame are expanded in the same proportion to obtain a portrait area PtEnsuring that the whole cutting area contains a complete portrait;
(25) for portrait region PtPreprocessing the portrait region PtScaling the long side to 224, scaling the short side equally, and supplementing 0 to the deficiency to obtain the RGB model input I with the size of 224X 3t;
(26) RGB model input ItObtaining a portrait mask output Y after the portrait segmentation model obtained in the step (1)t;
(27) Outputting Y to portrait masktPost-processing is carried out to optimize the result and obtain a binary portrait mask N corresponding to the original image framet(ii) a Outputting Y to portrait masktThe post-treatment specifically comprises the following steps: firstly, carrying out binarization processing on the image to obtain a binarized portrait mask, wherein the threshold value is 0.5; secondly, performing communication domain analysis on the portrait mask, removing the mistakenly-segmented regions, and reserving the maximum region of the portrait; the result is then scaled to the portrait area P in step (24)tThe size of (a); finally according to the portrait region PtThe cutting information of the image mask is compensated with 0 to obtain a binary image mask N corresponding to the original image framet;
(28) And (5) circulating the steps (23) to (27) until the image segmentation of the last frame is completed, and obtaining all portrait mask frames (N)1,N2,N3…)。
The invention adopts the traditional KCF tracking algorithm, and improves the running speed of the algorithm; the picture is taken as training data of the segmentation network, and compared with video data, the picture data is easier to acquire and label; designing a lightweight segmentation network, improving the running speed of the model and meeting the real-time requirement of a mobile terminal; therefore, the invention improves the running speed of the whole algorithm, and the tracking segmentation algorithm can meet the real-time performance requirement at the mobile terminal.
Claims (6)
1. A real-time portrait tracking and segmenting method is characterized by comprising the following steps:
(1) the training stage is used for training parameters of the human image segmentation model and performing off-line training on the segmentation network, and comprises two parts, namely data preprocessing and model training;
(2) in the prediction stage, picture sequence frames of a video are input into a portrait tracking segmentation algorithm, a portrait tracking frame is obtained through a KCF tracking algorithm, a portrait area is cut out according to the portrait tracking frame, the portrait area is preprocessed and input into a segmentation model, an output result is post-processed, a portrait segmentation mask corresponding to an input frame is obtained, the process is circulated according to the sequence of the video frames until a final video portrait mask sequence is obtained, and the process is finished.
2. A real-time human image tracking and segmenting method as claimed in claim 1, wherein in the step (1), the specific operation method is as follows:
(11) collecting different portrait data, and accurately marking a portrait area, wherein the background area is 0, and the portrait area is 1, so as to obtain a corresponding binary portrait mask;
(12) performing data enhancement processing on the training portrait data, then scaling the long edge of the image to 224, scaling the short edge in an equal ratio, and compensating for the deficiency to be 0 to be aligned to obtain an RGB input image I with the size of 224X 3xCarrying out the same scaling operation on the corresponding binary portrait mask to obtain a training portrait mask Iy;
(13) The method comprises the steps that Mobilenetv2 is used as a coding module of a segmentation network, and the whole coding module carries out 32-time down-sampling on an input image to obtain a feature map F; in the decoding module, a decoding mode similar to U-net is adopted to recover details of the feature diagram F, and after the size of the feature diagram is recovered to 56 × 56, 4 times of upsampling layers are directly utilized to obtain an output portrait mask Y with the size of 224 × 1;
(14) for output portrait mask Y and training portrait mask IyPerforming cross entropy loss function operation and Dice loss function operation to obtain a loss function L oss;
(15) the entire model is iterated using the penalty function L oss, updating the model parameters.
3. A method as claimed in claim 2, wherein in step (12) the data enhancement process comprises mirroring, rotation, luminance contrast transformation, affine transformation.
4. A real-time human image tracking and segmenting method as claimed in claim 1, wherein in the step (2), the specific operation method is as follows:
(21) performing frame dismantling processing on the video to obtain a picture sequence frame;
(22) inputting the first frame picture into a KCF tracking algorithm, manually marking a portrait frame needing to be tracked to initialize the KCF tracking algorithm, and taking the portrait frame as a portrait tracking frame B of the first frame1;
(23) Suppose the portrait tracking frame of the current frame is BtWhere t represents the position of the picture frame in the video, the KCF tracking algorithm tracks the frame B according to the portrait of the current frametTo predict the portrait tracking frame B of the next framet+1;
(24) Tracking the frame B according to the portrait obtained in the step (23)tSelf-adaptive clipping is carried out on the portrait area to obtain the portrait area Pt;
(25) For portrait region PtPreprocessing the portrait region PtScaling the long side to 224, scaling the short side equally, and supplementing 0 to the deficiency to obtain the RGB model input I with the size of 224X 3t;
(26) RGB model input ItObtaining a portrait mask output Y after the portrait segmentation model obtained in the step (1)t;
(27) Outputting Y to portrait masktPost-processing is carried out to optimize the result and obtain a binary portrait mask N corresponding to the original image framet;
(28) And (5) circulating the steps (23) to (27) until the image segmentation of the last frame is completed, and obtaining all portrait mask frames.
5. A method for real-time segmentation of human images according to claim 4 wherein in step (24) the adaptive cropping is: firstly, the aspect ratio of the portrait tracking frame is judgedOr the aspect ratio, if the aspect ratio or the aspect ratio is less than 0.5, the range of the short side is expanded to make the ratio reach 0.5; then, the width and the height of the portrait tracking frame are expanded in the same proportion to obtain a portrait area PtAnd the whole cutting area is ensured to contain a complete portrait.
6. A method as claimed in claim 4, wherein in step (27) the image mask is output with YtThe post-treatment specifically comprises the following steps: firstly, carrying out binarization processing on the image to obtain a binarized portrait mask, wherein the threshold value is 0.5; secondly, performing communication domain analysis on the portrait mask, removing the mistakenly-segmented regions, and reserving the maximum region of the portrait; the result is then scaled to the portrait area P in step (24)tThe size of (a); finally according to the portrait region PtThe cutting information of the image mask is compensated with 0 to obtain a binary image mask N corresponding to the original image framet。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010200344.2A CN111414860A (en) | 2020-03-20 | 2020-03-20 | Real-time portrait tracking and segmenting method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010200344.2A CN111414860A (en) | 2020-03-20 | 2020-03-20 | Real-time portrait tracking and segmenting method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111414860A true CN111414860A (en) | 2020-07-14 |
Family
ID=71493140
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010200344.2A Pending CN111414860A (en) | 2020-03-20 | 2020-03-20 | Real-time portrait tracking and segmenting method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111414860A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111932546A (en) * | 2020-08-20 | 2020-11-13 | 展讯通信(上海)有限公司 | Image segmentation model training method, image segmentation method, device, equipment and medium |
CN112487974A (en) * | 2020-11-30 | 2021-03-12 | 叠境数字科技(上海)有限公司 | Video stream multi-person segmentation method, system, chip and medium |
CN112529914A (en) * | 2020-12-18 | 2021-03-19 | 北京中科深智科技有限公司 | Real-time hair segmentation method and system |
CN113516672A (en) * | 2021-09-07 | 2021-10-19 | 北京美摄网络科技有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
CN114041767A (en) * | 2021-10-11 | 2022-02-15 | 宁波春建电子科技有限公司 | Heart rate detection method based on depth camera and millimeter wave radar |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062761A (en) * | 2017-12-25 | 2018-05-22 | 北京奇虎科技有限公司 | Image partition method, device and computing device based on adaptive tracing frame |
CN109977981A (en) * | 2017-12-27 | 2019-07-05 | 深圳市优必选科技有限公司 | Scene analysis method based on binocular vision, robot and storage device |
CN110399847A (en) * | 2019-07-30 | 2019-11-01 | 北京字节跳动网络技术有限公司 | Extraction method of key frame, device and electronic equipment |
CN110490858A (en) * | 2019-08-21 | 2019-11-22 | 西安工程大学 | A kind of fabric defect Pixel-level classification method based on deep learning |
-
2020
- 2020-03-20 CN CN202010200344.2A patent/CN111414860A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062761A (en) * | 2017-12-25 | 2018-05-22 | 北京奇虎科技有限公司 | Image partition method, device and computing device based on adaptive tracing frame |
CN109977981A (en) * | 2017-12-27 | 2019-07-05 | 深圳市优必选科技有限公司 | Scene analysis method based on binocular vision, robot and storage device |
CN110399847A (en) * | 2019-07-30 | 2019-11-01 | 北京字节跳动网络技术有限公司 | Extraction method of key frame, device and electronic equipment |
CN110490858A (en) * | 2019-08-21 | 2019-11-22 | 西安工程大学 | A kind of fabric defect Pixel-level classification method based on deep learning |
Non-Patent Citations (1)
Title |
---|
FANYI XIAO ET AL: "《Track and Segment: An Iterative Unsupervised Approach for Video Object Proposals》", 《IEEE》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111932546A (en) * | 2020-08-20 | 2020-11-13 | 展讯通信(上海)有限公司 | Image segmentation model training method, image segmentation method, device, equipment and medium |
CN112487974A (en) * | 2020-11-30 | 2021-03-12 | 叠境数字科技(上海)有限公司 | Video stream multi-person segmentation method, system, chip and medium |
CN112529914A (en) * | 2020-12-18 | 2021-03-19 | 北京中科深智科技有限公司 | Real-time hair segmentation method and system |
CN112529914B (en) * | 2020-12-18 | 2021-08-13 | 北京中科深智科技有限公司 | Real-time hair segmentation method and system |
CN113516672A (en) * | 2021-09-07 | 2021-10-19 | 北京美摄网络科技有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
CN113516672B (en) * | 2021-09-07 | 2022-02-25 | 北京美摄网络科技有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
CN114041767A (en) * | 2021-10-11 | 2022-02-15 | 宁波春建电子科技有限公司 | Heart rate detection method based on depth camera and millimeter wave radar |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111414860A (en) | Real-time portrait tracking and segmenting method | |
Kumar et al. | Image smog restoration using oblique gradient profile prior and energy minimization | |
CN106709964B (en) | Sketch generation method and device based on gradient correction and multidirectional texture extraction | |
CN110969589A (en) | Dynamic scene fuzzy image blind restoration method based on multi-stream attention countermeasure network | |
CN110796662B (en) | Real-time semantic video segmentation method | |
CN116797488A (en) | Low-illumination image enhancement method based on feature fusion and attention embedding | |
CN108989731B (en) | Method for improving video spatial resolution | |
CN108986185B (en) | Image data amplification method based on deep learning | |
CN112233129A (en) | Deep learning-based parallel multi-scale attention mechanism semantic segmentation method and device | |
CN115393396B (en) | Unmanned aerial vehicle target tracking method based on mask pre-training | |
CN112767280B (en) | Single image raindrop removing method based on loop iteration mechanism | |
CN112184585A (en) | Image completion method and system based on semantic edge fusion | |
CN110992374A (en) | Hair refined segmentation method and system based on deep learning | |
CN113902925A (en) | Semantic segmentation method and system based on deep convolutional neural network | |
CN111652231B (en) | Casting defect semantic segmentation method based on feature self-adaptive selection | |
CN114627269A (en) | Virtual reality security protection monitoring platform based on degree of depth learning target detection | |
CN113962905A (en) | Single image rain removing method based on multi-stage feature complementary network | |
CN111507279B (en) | Palm print recognition method based on UNet + + network | |
CN116895037A (en) | Frame insertion method and system based on edge information and multi-scale cross fusion network | |
CN115457448B (en) | Intelligent extraction system for video key frames | |
CN108550119B (en) | Image denoising method combined with edge information | |
CN112200751B (en) | Image enhancement method | |
CN112733714B (en) | VGG network-based automatic crowd counting image recognition method | |
CN114627139A (en) | Unsupervised image segmentation method, unsupervised image segmentation device and unsupervised image segmentation equipment based on pixel feature learning | |
CN115393491A (en) | Ink video generation method and device based on instance segmentation and reference frame |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200714 |
|
RJ01 | Rejection of invention patent application after publication |